Model training method and apparatus based on gradient boosting decision tree

Abstract

Disclosed are a model training method and apparatus based on gradient boosting decision tree (GBDT). A GBDT algorithm flow is divided into two stages. In the first stage, labeled samples are obtained from a data domain of a service scenario similar to a target service scenario to sequentially train several decision trees, and training residual generated after the training in the first stage is determined; in the second stage, labeled samples are obtained from a data domain of the target service scenario, and several decision trees continue to be trained based on the training residual. Finally, a model applied to the target service scenario is actually obtained by integrating the decision trees trained in the first stage with the decision trees trained in the second stage.

Description

BACKGROUNDTechnical Field[0001]Implementations of the present specification pertain to the field of information technologies, and in particular, to a model training method and apparatus based on gradient boosting decision tree (GBDT).Description of the Related Art[0002]Machine learning is an important branch of computer technologies. Many machine learning methods require proper training based on relevant data in particular context. In many cases, when a prediction model that is designed for applying to a certain service scenario needs to be trained, a large amount of data needs to be obtained from a data domain of this service scenario for labeling, so as to obtain labeled samples for model training. If there are only a small number of labeled samples, it is usually impossible to obtain a qualified model. It should be noted that a data domain of a certain service scenario is actually a set of service data generated based on the service scenario.[0003]However, in practice, there may ...

AI Technical Summary

Benefits of technology

The present patent provides a method and apparatus for training a model using the GBDT algorithm. The technique improves training data sufficiency by dividing the training process into two stages. In the first stage, decision trees are trained using labeled samples from a data domain similar to the target service scenario, and the training residual is determined. In the second stage, decision trees are trained using labeled samples from the target service scenario and the training residual from the first stage. The models trained in the first and second stages are then integrated to obtain the final model applied to the target service scenario. This approach allows for efficient training even when the data in the target scenario is insufficient. The patent improves machine learning technology and enables earlier and effective training of various prediction models, even if specific, contextual data has not been accumulated to a level sufficient for traditional model training purposes.

Problems solved by technology

If there are only a small number of labeled samples, it is usually impossible to obtain a qualified model.

However, in practice, there may be insufficient accumulation of data in some service scenarios.

Consequently, it can be impossible to obtain enough labeled samples from the data domain of a certain service scenario when a model designed for applying to the service scenario needs to be trained, and no qualified model can be obtained.

Images